I wonder if there have been cases where a pilot had to cut fuel before the computer could detect anything abnormal? I do realize that defining "abnormal" is the hardest part of this algorithm.
I wonder if there have been cases where a pilot had to cut fuel before the computer could detect anything abnormal? I do realize that defining "abnormal" is the hardest part of this algorithm.
As a software engineer myself I think it's interesting that we feel software is the true solution when we wouldn't accept that solution ourselves. For example typically in a company you do code reviews and have a release gating process but also there's some exception process for quickly committing code or making adjustments when theres an outage or something. Could you imagine if the system said "hey we aren't detecting an outage, you sure about that? why don't you go take a walk and get a coffee, if you still think there's an outage in 15 minutes from now we will let you make that critical change".
In safety-critical engineering, you generally either automate things fully (i.e. to exceed human capabilities in all situations, not just most), or you keep them manual. Half-measures of automation kill people.
I wonder if they could have buttons that are about the situation rather than the technical action. Have a fire response button. Or a shut down on the ground button.
But it does seem like half measure automation could be a contributing factor in a lot of crashes. Reverting to a pilot in a stressful situation is a risk, as is placing too much faith in individual sensors. And in a sense this problem applies to planes internally or to the whole air traffic system. It is a mess of expiring data being consumed and produced by a mix of humans and machines. Maybe the missing part is good statistical modelling of that. If systems can make better predictions they can be more cautious in response.
warn then shut off
Else
Shut off immediately
EndOverride warning time by toggling again.